ElasticXL SYNC redundancy

[Wolfgang]

We want to have redundancy for the SYNC connection in a ElasticXL cluster (2 appliances).

SYNC is a BOND and it's possible to add additional interface to this BOND. We want to use one connection direct connected between the two appliances and another one via switches. With this we can't configure the BOND as LACP-channel, because both ends of the BOND are not on the same devices.

But how about configuring the BOND as active/backup ? Not both are active but if one is failing the second is used.

I don't know which BOND mode is used by default for the SYNC Bond.

[Chris_Atkinson · Accepted solution]

I'm fairly certain the Sync-bond is active-backup by default.

You may check it using: 

[Expert@EXL-s01-01:0]# clish -c "show configuration" | grep 1024

[Bob_Zimmerman · Accepted solution]

It definitely is active-backup by default.

[Expert@DallasticXL-s01-01:0]# clish -c "show configuration" | grep 1024
add bonding group 1024
set bonding group 1024 mode active-backup
set bonding group 1024 primary eth1-Sync
set bonding group 1024 xmit-hash-policy layer2
add bonding group 1024 interface eth1
add bonding group 1024 interface eth1-Sync

Note that active-backup doesn't test the alternate paths. If it's active on the direct-cabled path and that fails for some reason, you might discover the switched path failed some time ago without being noticed.

I personally prefer round-robin for this reason, though with round-robin, you can end up in a situation where one member's link to the switch has failed. The symptom of this is the member with the bad link can reliably send traffic to other members, but it can't reliably receive traffic from them.

[Wolfgang]

Thanks @Bob_Zimmerman and @Chris_Atkinson for your suggestions. I really understand the behaviour with active/backup and the problematic two links from gateway to switch. I prefer a LACP bond, with LACP a probing is done and it’s known which link is working. But our customer prefer to use the mentioned links.